Conditional random field-based offline map matching for indoor positioning

  1. Bataineh, Safaa Ahmed Hijris
Dirigida por:
  1. Enrique Onieva Caracuel Director
  2. Alfonso Bahillo Director

Universidad de defensa: Universidad de Deusto

Fecha de defensa: 07 de julio de 2017

Tribunal:
  1. Rubén Mateo Lorenzo Toledo Presidente/a
  2. Diego López de Ipiña González de Artaza Secretario
  3. Fernando Javier Álvarez Franco Vocal

Tipo: Tesis

Resumen

The indoor localization is of great importance for many context-aware applications. Although many indoor localization systems have been proposed, this area is still open for research; new methods are needed to overcome the tradeoffs between cost, accuracy, robustness and complexity. Using the map information is one of the methods to improve the feasibility and the accuracy of the localization systems. Unlike many systems that fuse the map information with other measurements in the same algorithm, the proposed system separates the map matching step from the basic localization process which makes it compatible with various different localization systems. The proposed map matching algorithm is independent of the positioning system implementation and it can be applied offline and coupled to different indoor positioning systems that use the coordinates to represent their output trajectory. For this purpose a platform for indoor map matching was created using Matlab. The platform uses the building plans presented in CAD files to automatically generate semantic maps suitable for map matching. A map matching algorithmbased on the Conditional Random Field (CRF) probabilistic model is developed to match trajectories obtained by indoor positioning systems with the map. A ground truth was constructed and actual input trajectories were obtained by real experiments and by simulations; the output map-matched trajectories were tested using the ground truths and the accuracy of the output corrected trajectories were measured. The behavioral information of pedestrians has been used to enhance the results by smoothing the output trajectories. The parameters were adjusted in order to obtain the best possible results. The algorithm can be developed in the future by using different map models and more features.